search

NVIDIA / nvbandwidth

A tool for bandwidth measurements on NVIDIA GPUs.
Apache License 2.0
315 stars 30 forks source link

Waiving device_to_device_memcpy_write_ce for RTX 2080 Ti #13

Closed thisiszy closed 11 months ago

thisiszy commented 11 months ago

I ran the benchmark on a single node with two NVIDIA GeForce RTX 2080 Ti GPUs. I tested device_to_device_memcpy_write_ce, however, the program waived the test. 

$ ./nvbandwidth -t device_to_device_memcpy_write_ce
nvbandwidth Version: v0.2
Built from Git version: 6cefdda

NOTE: This tool reports current measured bandwidth on your system.
Additional system-specific tuning may be required to achieve maximal peak bandwidth.

CUDA Runtime Version: 12010
CUDA Driver Version: 12000
Driver Version: 525.125.06

Device 0: NVIDIA GeForce RTX 2080 Ti
Device 1: NVIDIA GeForce RTX 2080 Ti

Waiving device_to_device_memcpy_write_ce.

host_to_device_bidirectional_memcpy_ce and host_to_device_memcpy_ce tests are fine.

deepakcu commented 11 months ago

The test gets waived because the peer to peer mapping failed. Peer-to-peer mapping over PCIe is likely not supported for your configuration.

thisiszy commented 11 months ago

Aha, I see. I will ask my cluster admin. Thank you!